Micron and Western Digital Rewrite Technology News: Will RAM and Drive Prices Fall Again?

Micron and Western Digital entered 2026 with memory and storage supply unusually tight, despite years of falling consumer hardware costs. That reversal has made component pricing a major technology news story. The immediate question is no longer when the next routine discount arrives. It is whether artificial intelligence has changed the market enough to delay a meaningful decline.

A question about falling RAM and drive prices reached Zhihu’s Hot List on August 11, 2026. The discussion reflects a much broader concern among PC builders, corporate buyers, and data center operators. DRAM, NAND flash, enterprise SSDs, and high-capacity hard drives have all faced pressure during the same cycle.

The simple answer is that some retail products will receive temporary discounts. However, a broad return to earlier price levels is unlikely during the next several quarters. Micron expects tight DRAM and NAND conditions beyond 2026, while Western Digital has described strong demand for high-capacity drives.

This cycle also differs from earlier shortages. Cloud providers are signing longer supply agreements, manufacturers are favoring AI-related products, and new fabrication capacity takes years to become productive. Weak consumer demand can slow increases, but it does not immediately release the capacity needed for a lasting price collapse.

Technology News Has Shifted From Chip Shortages to Allocation

The most important change is not that memory became harder to buy. Manufacturers changed which customers and products receive scarce capacity first.

TrendForce said conventional DRAM contract prices were expected to rise 58% to 63% during the second quarter of 2026. It projected a 70% to 75% increase for NAND flash during the same period. Those estimates followed even larger DRAM increases during the first quarter.

The pace moderated during the third quarter, but the direction did not reverse. TrendForce later projected conventional DRAM contract prices to rise another 13% to 18%. It expected NAND flash contract prices to increase 10% to 15%.

Those figures describe contract markets rather than every retail product. A specific SSD can still go on sale because a retailer has excess stock. A memory kit can also fall after a short-lived inventory imbalance.

Contract prices matter because they shape the replacement cost behind future consumer products. A retailer may discount old inventory, but the next shipment can arrive under less favorable terms. This creates brief bargains without establishing a durable downward trend.

The underlying mechanism starts with high-bandwidth memory, or HBM, which feeds data to AI accelerators at very high speeds. HBM is not interchangeable with desktop DDR5. However, it competes for investment, engineering resources, production equipment, and advanced packaging.

Memory manufacturers have strong reasons to prioritize it. AI customers buy large volumes, accept long commitments, and require premium products. Conventional DRAM customers often face greater price sensitivity and shorter purchasing cycles.

NAND flash is experiencing a related pull from enterprise SSD demand. AI training creates large datasets, while inference produces logs, embeddings, cached information, and generated media. Enterprises must retain much of that data after accelerator processing ends.

TrendForce’s memory forecast connected the second-quarter increases to AI server demand and long-term purchasing agreements. These agreements make supply more predictable for cloud operators. They also reduce the uncommitted inventory available to other buyers.

The hard-drive market adds another layer. HDDs do not use NAND, but hyperscale data centers still rely on them for economical mass storage. AI can increase demand for both fast flash and lower-cost persistent capacity.

Western Digital said virtually every AI workload creates information that must be stored persistently. Its April 30 results highlighted demand from training, inference, agentic systems, and physical AI. The company framed HDDs as the cost-efficient destination for much of that growing data.

This matters because buyers cannot assume that SSD growth will automatically weaken hard-drive demand. The two technologies increasingly occupy different positions in the same infrastructure. SSDs handle latency-sensitive work, while HDDs retain large volumes more economically.

The result is simultaneous pressure across categories that once followed more independent cycles. DRAM feeds processors. NAND supports fast storage. HDDs preserve expanding datasets. AI infrastructure can consume all three without treating them as direct substitutes.

That synchronized demand explains why the current story feels different to consumers. A PC builder cannot easily escape higher memory costs by switching storage formats. An enterprise buyer cannot replace every high-capacity HDD with flash without changing its budget and architecture.

It also explains why one weak consumer quarter offers limited relief. Manufacturers can redirect output toward stronger enterprise orders. Falling notebook demand might reduce one source of pressure, yet it does not cancel contracted data center requirements.

AI Buyers Are Pressuring Consumers and Smaller Businesses

The allocation shift places the greatest pressure on buyers who lack long contracts, large orders, or strategic value to manufacturers.

Hyperscalers plan infrastructure purchases years ahead. They negotiate directly with component makers and can commit to substantial volumes. A small computer manufacturer, local business, or individual buyer operates much closer to the spot and retail markets.

That difference changes who absorbs volatility. Large customers can secure supply while accepting negotiated price ranges. Smaller buyers face whatever inventory remains after major contracts, product qualifications, and manufacturing priorities are settled.

Micron’s fiscal third-quarter results, released June 24, provide a clear signal. The company said the value of memory had become more strategic during the AI era. It also described record investment in technology, products, and supply.

The company’s prepared remarks went further. Micron expected supply and demand for DRAM and NAND to remain tight beyond calendar 2026. That assessment came from a major producer with direct visibility into customer orders and planned output.

Micron’s quarterly results do not guarantee uninterrupted increases. Corporate forecasts can change, and customers sometimes double-order during shortages. Still, the statement weakens the case for a rapid return to old price levels.

PC manufacturers face a difficult choice. They can raise system costs, reduce memory or storage specifications, accept lower margins, or delay promotions. Premium models usually protect margins better, so affordable systems often receive the most visible compromises.

A notebook that previously included generous memory may ship with less capacity. A desktop vendor can use a smaller SSD while keeping other headline specifications unchanged. Buyers then pay indirectly through future upgrades or tighter local storage.

Small businesses feel the same pressure at a different scale. A workstation refresh becomes harder to budget when RAM, SSDs, and backup drives rise together. Delaying purchases can help, but older systems bring reliability and support risks.

Cloud customers do not escape the cycle either. Infrastructure providers eventually pass hardware and financing costs into service design. They can change discounts, commitment structures, storage classes, or resource availability without presenting a simple component surcharge.

Developers may therefore encounter the shortage through constrained hardware choices rather than a retail receipt. Memory-intensive local models become less attractive. Hosted inference can also become harder to optimize when providers prioritize their own accelerator utilization.

Knowledge workers face a quieter version of the problem. Larger local files, recorded meetings, generated media, and AI indexes consume storage steadily. Replacing a laptop solely because its memory cannot be upgraded becomes a more consequential decision.

The effect is especially sharp for users who postponed upgrades during previous declines. Historically, waiting often delivered more capacity for less money. During a supply-constrained upcycle, waiting can instead mean choosing between an older system and a less favorable replacement.

This does not mean everyone should buy immediately. Panic purchasing can amplify shortages and leave buyers holding unnecessary inventory. The more useful response is to identify which capacity is essential and which upgrades remain optional.

For a PC builder, adequate RAM for the intended workload matters more than chasing the highest speed grade. For storage, separating active files from archival data can reduce the need for premium flash. Businesses can also schedule purchases before a critical failure forces an urgent order.

The key pressure is time. Large customers gain leverage by planning early and committing volume. Consumers and smaller organizations usually make decisions later, when supply conditions have already shaped the available products.

The Old Memory Cycle Is Competing With Structural AI Demand

Prices can fall again, but the traditional oversupply cycle must first overcome demand that is broader, better financed, and more heavily contracted.

Memory has always been cyclical. Suppliers expand capacity when prices and margins are strong. New production eventually reaches the market, inventories rise, and buyers gain leverage. Manufacturers then reduce output until supply and demand rebalance.

That history supports the argument for eventual declines. No hardware category rises forever. Higher costs discourage purchases, encourage efficiency, and attract investment in additional production.

The disagreement concerns timing and depth. A normal cycle might produce meaningful relief after a few quarters. The current cycle includes constraints that can keep the market tight even while consumer demand weakens.

New fabrication plants require long construction and qualification schedules. Equipment installation does not create saleable output immediately. Manufacturers must tune processes, improve yields, and qualify products with major customers.

Technology transitions can also consume nominal capacity gains. More advanced DRAM and higher-layer NAND increase bit output over time, but early production can carry lower yields. HBM additionally requires complex assembly and advanced packaging.

Long-term agreements reduce another source of cyclical volatility. When cloud customers reserve future supply, producers receive stronger demand visibility. They can expand more deliberately instead of building speculative capacity for an uncertain consumer rebound.

That discipline benefits manufacturers after painful historical downturns. It also means buyers should not expect producers to flood the market at the first sign of strong margins. Companies have incentives to avoid recreating severe oversupply.

Gartner has described the problem as “memflation,” a period when constrained memory supply lifts costs across technology products. Its shortage outlook extends the imbalance into the second half of 2027.

The forecast deserves caution because industry projections can miss sudden changes. A recession could weaken device demand. AI spending could slow. New capacity might ramp faster than expected, while efficiency improvements could reduce memory required per workload.

Yet those possibilities must offset several active demand sources. AI training needs HBM and DRAM. Inference expands server memory and storage requirements. Smartphones and PCs continue adopting more memory, even when unit shipments soften.

On-device AI adds another demand channel. Models running on laptops and phones need enough local memory to operate without constant cloud access. The feature can increase memory content per device even if total device sales remain modest.

Storage demand has a similar multiplier. AI systems do not only consume training data once. They create checkpoints, vector indexes, cached responses, telemetry, synthetic datasets, and user-generated outputs.

A vector index organizes numerical representations so software can retrieve related information efficiently. It can improve search and retrieval, but it also creates another persistent storage layer beside the original documents.

Companies can delete or compress some of this material. Better models can also use smaller caches. However, lower computing costs often encourage more usage, which can offset efficiency gains.

The primary contest is therefore clear. The old cycle predicts that investment and weak demand will eventually create oversupply. The structural-demand view says AI customers will absorb new capacity before consumers see major relief.

Evidence in 2026 favors the structural-demand side over the short term. Micron sees continued tightness. Western Digital reports persistent high-capacity storage demand. SK hynix has warned that customer requirements are exceeding available supply.

The longer-term outcome remains less certain. Large capacity additions scheduled for later years can change the balance. AI infrastructure spending must also produce enough economic value to justify repeated expansion.

If AI revenue disappoints, cloud providers can slow new orders faster than fabrication plants can adjust. That would recreate a classic oversupply setup. Prices could then fall sharply rather than decline smoothly.

However, that scenario requires more than consumers buying fewer PCs. It requires a meaningful change in enterprise procurement, cloud capital spending, or contracted demand. Without one of those shifts, retail weakness mainly slows increases.

Why SSD and HDD Prices Will Not Move Together

A single answer for “drive prices” hides two supply chains, several product classes, and very different customer priorities.

An SSD stores data in NAND flash, while an HDD records it on rotating magnetic disks. Both provide persistent storage, but their economics and manufacturing constraints differ substantially.

Client SSDs serve laptops, desktops, and game systems. Enterprise SSDs use stricter endurance, reliability, and performance requirements. Demand for one category can influence NAND allocation without producing identical retail movements.

Consumer SSD prices can therefore soften before the broader NAND market does. Retailers may clear older models, and suppliers can redirect certain flash grades. Competition among controller and module vendors also creates short-lived promotions.

Those promotions do not prove that the underlying cycle has ended. A durable decline requires lower replacement costs across several purchasing rounds. Buyers should distinguish a discounted product from a falling market.

Enterprise SSD pressure can remain strong because AI inference needs fast access to large datasets. Smaller, high-performance drives may face additional constraints when they require scarce DRAM components alongside NAND.

Hard drives follow another path. The market is concentrated, and high-capacity nearline drives require specialized manufacturing. Nearline HDDs are designed for data centers that need large, economical storage pools with frequent access.

Western Digital’s storage results tied persistent HDD demand directly to AI-created data. That claim aligns with the architecture used by major cloud operators.

Seagate has made a similar case. Its fiscal third-quarter announcement said AI applications amplify data creation and support sustained storage demand. The company also emphasized its mass-capacity products rather than consumer drives.

Seagate’s capacity strategy relies partly on heat-assisted magnetic recording. HAMR heats a tiny disk area during writing, allowing manufacturers to store more data on each platter.

Higher density can lower the physical resources needed per unit of storage. It does not instantly produce unlimited drives. New platforms still need manufacturing ramps, customer testing, and reliable yields.

This creates an important split for buyers. Smaller consumer HDDs can experience weak demand because SSDs dominate many personal computers. High-capacity data center drives can remain constrained at the same time.

External desktop drives may occasionally become attractive because retail demand is uneven. Enterprise-grade models can follow a firmer trajectory. Capacity, warranty, workload rating, and distribution channel all affect the result.

The question “Will hard-drive prices fall?” therefore needs a use case. A home user seeking archival storage may find promotions. A company expanding a large backup cluster faces a different supply environment.

SSD buyers face the same need for specificity. Entry-level QLC drives store more bits per cell and can reduce manufacturing cost. They may fall sooner than high-end enterprise models with demanding endurance targets.

QLC means four bits are stored in each flash cell. The design increases density, though performance and endurance depend heavily on the controller, workload, and available cache.

Older interfaces can also behave unpredictably. A mature SATA SSD is not automatically cheaper if suppliers reduce production. Scarcity can keep legacy products expensive even after faster alternatives become common.

DRAM shows a comparable pattern. DDR4 can rise when manufacturers migrate capacity toward newer DDR5 and HBM products. An older standard may become more expensive because fewer suppliers want to produce it.

Buyers should avoid treating age as a guarantee of lower cost. The relevant question is whether production remains plentiful relative to active demand. A discontinued or deprioritized product can resist the usual depreciation curve.

The most credible path to lower SSD prices begins with weaker enterprise NAND orders or faster supply growth. HDD relief requires expanded high-capacity output, slower hyperscale demand, or a clear reduction in purchasing commitments.

Consumer promotions will appear before either condition becomes obvious. They are useful buying opportunities, but they are poor evidence for a market-wide reversal.

What Could Still Break the High-Price Narrative

The strongest argument against prolonged tightness is that every shortage encourages conservation, substitution, and investment until demand expectations become too optimistic.

Memory producers and drive manufacturers have an interest in describing AI demand as durable. Strong demand supports investment plans, customer commitments, and favorable margins. Their forecasts should not be treated as neutral guarantees.

The first uncertainty concerns AI spending itself. Cloud companies are investing heavily, but the long-term return remains unsettled. If customers resist higher service costs, infrastructure growth can slow.

AI developers are also improving efficiency. Quantization reduces the numerical precision used to store and process model parameters. Smaller representations can lower memory use while preserving acceptable output quality.

Caching, model routing, and retrieval design can reduce unnecessary work. Companies can send simple requests to smaller models and reserve large systems for difficult tasks. That reduces the average memory and computing requirement per request.

Data management can improve as well. Organizations can delete redundant checkpoints, compress archives, and set shorter retention periods for low-value telemetry. Storage consumption does not have to match raw data creation.

The counterargument is that efficiency often expands usage. Lower resource requirements make new products economical, which attracts more users and generates more data. Total demand can rise even while each task becomes cheaper.

The second uncertainty concerns new supply. Samsung, SK hynix, Micron, Kioxia, Sandisk, and Chinese manufacturers all have incentives to capture demand. Additional output can arrive through new facilities, process improvements, and higher-density products.

Capacity announcements must be interpreted carefully. A planned facility is not the same as qualified production. Construction, equipment delivery, yield improvement, and customer approval can separate an announcement from meaningful output.

Chinese DRAM expansion presents another possible pressure valve. It can increase competition in mainstream products, especially outside markets affected by trade restrictions or customer qualification rules. However, geopolitical limits may prevent that supply from balancing every region.

The third uncertainty is inventory. Shortages encourage customers and distributors to order more than they immediately need. This behavior can make true demand difficult to measure.

Once buyers believe supply is improving, they can stop replenishing and consume existing inventory. Orders then fall faster than end-user demand. Memory markets have repeatedly turned on that inventory correction.

The fourth uncertainty is consumer resistance. TrendForce’s third-quarter outlook already indicated that weaker consumer conditions were moderating the rate of increases. Affordability limits can force PC and smartphone makers to cut specifications or shipment targets.

A slower increase is not a decline. Still, it is often the first stage of a transition. If weak device sales persist while new output arrives, suppliers lose pricing power.

The fifth uncertainty is product substitution. Enterprises can place cold data on tape, optimize compression, or reconsider retention policies. Consumers can use network storage, cloud services, or smaller local drives.

None of these options replaces every use case. They can still reduce marginal demand, and marginal demand often determines whether a tight market becomes balanced.

SK hynix CEO Kwak Noh-jung presented the opposite view in July. He said the industry was approaching its most severe shortage in 2027 and expected demand to exceed supply beyond 2030. The comments were reported in a supply warning.

That forecast captures the bullish case for memory producers, but its long horizon creates substantial uncertainty. AI architectures, trade policy, capital spending, and manufacturing yields can all change before 2030.

Buyers should therefore reject two extremes. Prices are not guaranteed to keep rising without interruption. They are also unlikely to return quickly to earlier lows merely because consumer demand weakens.

The more defensible expectation is uneven movement. Retail discounts will coexist with firm contract pricing. Some consumer SSDs will soften while enterprise products remain constrained. Older DRAM can stay expensive even when newer products gain output.

Three Signals Will Show When Prices Can Fall Again

A real decline will require evidence from producer forecasts, customer contracts, and inventory, not merely a weekend promotion.

The first signal is a change in manufacturer guidance. Watch whether Micron stops describing both DRAM and NAND as tight beyond the current calendar year. Similar language from Samsung and SK hynix would strengthen that signal.

Producer guidance matters because these companies see order schedules, factory yields, and customer negotiations. One cautious statement is insufficient. A coordinated shift across several suppliers would suggest the market balance is genuinely changing.

If guidance remains tight while retail products go on sale, the discount probably reflects channel inventory. If manufacturers begin warning about excess supply, the old memory cycle is returning.

The second signal is weaker long-term contracting from hyperscalers and enterprise customers. Reserved capacity currently limits what can reach other markets. Shorter commitments or renegotiated volumes would release supply and reduce producer visibility.

Listen for references to cancellations, delayed data center construction, lower infrastructure spending, or unused capacity. These events would weaken the structural-demand argument.

The absence of new long-term agreements also matters. Customers do not need to cancel existing contracts for the market to loosen. They can simply avoid extending them into later periods.

For HDDs, monitor order coverage at Western Digital and Seagate. If the companies move from heavily committed production toward shorter lead times, high-capacity storage is becoming easier to obtain.

The third signal is a sustained inventory increase accompanied by lower contract prices. Inventory alone can be misleading because companies build stock ahead of launches or anticipated shortages.

The meaningful combination is rising inventory, weaker orders, and consecutive contract-price declines. That pattern would show that supply is reaching buyers faster than demand consumes it.

Spot prices can move earlier, but they are volatile. A brief decline may reflect traders reducing positions rather than a durable improvement. Contract prices provide a stronger confirmation because they affect larger purchasing cycles.

For consumers, this framework leads to a practical answer. RAM and drive prices will eventually decline, but the market has not established the conditions for a broad return to previous lows. AI-related demand and deliberate supply allocation remain too influential.

Buying decisions should follow need rather than predictions of the exact bottom. Replace unreliable hardware before it threatens important data. Purchase essential workstation memory when the workload is known, but avoid speculative stockpiling.

When an upgrade is optional, track several comparable products over time. A genuine downward trend should appear across brands and retailers, not only one clearance listing.

Businesses should separate immediate capacity needs from flexible expansion. They can secure critical components while delaying nonessential upgrades. Longer planning windows also reduce exposure to emergency purchases.

Developers can reduce pressure through efficient models, selective caching, and tiered storage. Those choices will not reverse the global cycle, but they can lower the amount of scarce hardware each project requires.

The next decisive technology news will not be another isolated sale. It will be a supplier acknowledging weaker orders, a hyperscaler reducing commitments, or contract prices falling across multiple quarters.

Until those signals appear together, expect a divided market. Some products will get cheaper, especially during promotions and transitions. The broad cost of memory and storage is unlikely to retrace its earlier path quickly.

The question for buyers is therefore not whether prices can ever fall. They can. The useful question is whether waiting creates more savings than risk for a specific workload, system, or storage plan.